Oscar Dowson

I've updated the description of this issue with my notes about a JuMP linter.

cc @RobbyBP: this is the issue I was talking about this morning

Obligatory link to where I automatically try to reset the optimizer in SDDP https://github.com/odow/SDDP.jl/blob/2fd1e8d89b562df97fa23597b36b6c3a18c185a5/src/algorithm.jl#L295-L322

I'll also leave a link to this section of the TimeFold docs, which discuss testing: https://docs.timefold.ai/timefold-solver/latest/constraints-and-score/score-calculation#constraintStreamsTesting

This tooling should also have some way of tracking/profiling where JuMP spends its time: https://github.com/jump-dev/JuMP.jl/issues/3910 Pyomo has a built in profiler: see pg 129 of https://www.osti.gov/servlets/purl/1771935 The other place to...

We could detect and recommend `direct_model` if you don't use bridges and the optimizer supports in the incremental interface.

We could add a timer to each macro that runs. This would capture most useful timing information. It'd miss `x += y` but that's not the end of the world.

@joaquimg: feeding into the "dual feasibility" idea, check convergence of a NLP https://discourse.julialang.org/t/the-solution-deteriorates-if-the-domain-of-decisions-is-improperly-set-when-using-ipopt/126743/3

Actually the underlying issue of https://discourse.julialang.org/t/the-solution-deteriorates-if-the-domain-of-decisions-is-improperly-set-when-using-ipopt/126743 would be fixed by checking if the function and gradient is defined at the starting point.